基于l_pBTV正则化的图像超分辨率重建算法

Super-resolution Image Reconstruction Algorithm Based on l_pBTV Regularization

针对双边全变分正则化(BTV)边缘细节处理时出现过平滑的问题,结合自然图像梯度稀疏先验模型,给出一种基于双边全变分算子的新型l_pBTV(0<p<1)正则子。将其应用于图像超分辨率重建处理,采用迭代加权最小二乘法(IRLS),将l_pBTV非凸优化问题转化为重赋权的BTV优化问题。实验结果表明,该文算法能通过调整p值大小取得最优峰值信噪比(PSNR)和结构相似度(SSIM),可更有效地抑制噪声并保留图像的边缘细节,获得较好的图像重建效果,能有效提高图像质量。

In order to solve the problem of Bilateral Total Variation （BTV） which tends to produce excessive smoothing, an improved BTV algorithm called lp BTV （0〈p 〈 1 ） is proposed which based on traditional BTV regularization and combined with the natural image sparse gradient prior model. By using Iteratively Reweighted Least Squares （IRLS） algorithm, the/pBTV non-convex optimization problem is transformed into a reweighted BTV optimization problem in super-resolution image reconstruction. Experimental results show that the optimal Peak Signal to Noise Ratio （PSNR） and the Structural Similarity （SSIM） can be obtained by adjusting the p value. Compared to the existing BTV improved algorithms the proposed algorithm can remove noise and preserve edge details of the image more effectively. It can obtain better reconstruction effect and improve the quality of the image more effectively.